Study On Video-Based Vehicle Detection And License Plate Location

Advanced Traffic Management Systems (ATMS) has attracted more and more attention as an important research direction of Intelligent Transportation Systems (ITS). The intelligent road vehicle monitoring and recording system (hereinafter referred to as vehicle monitoring system) as a typical application example of the ATMS contains three basic problems:image compression coding, video vehicle detection and license plate location. While image compression coding can effectively reduce storage space and enhance network transmission efficiency of the images. Vehicle detection can provide various useful parameters of vehicles and is the basic link of the vehicle monitoring system; video-based vehicle detection technology with easy implementation and good portability becomes the current mainstream technology. License plate location is the first step of automatic vehicle license plate recognition module and meanwhile can be an approach for vehicle detection. Based on the above problems, this paper does the following researches:Wavelet analysis has features of good time-frequency adaptive and multi-resolution analysis, and image compression coding algorithms based on wavelet analysis for characteristics of high compression ratio and high compression speed. As spatial orientation trees are used to describe zero-tree structures in the image wavelet coefficients, image coding complexity is reduced in the set partitioning in hierarchical trees (SPIHT) algorithm, thus SPIHT algorithm using lifting 9/7 wavelet is adopted for still image decomposition in this paper. Image sequences in the vehicle intelligent observation system have fixed background and obvious vehicle objects, certain researches are done in this paper for image sequences coding using moving object detection and motion estimation theories.Common video-based vehicle detection methods such as optical flow method, frame difference method and background difference method can be improved for better accuracy and rapidity. An improved method based on combining frame difference method and background difference method is adopted in this paper, which uses sampling frame difference method for pre-detection and uses background difference method for accurate detection of the vehicle targets. The proposed method not only develops the rapidity of the frame difference method, but also keeps the accuracy of the background difference method, and can restrain influences of the noises effectively. Finally, shadows of the vehicle targets are removed by edge detection and color analysis.Compared with other parts of the car, the area of license plate has rich texture, color information and high contrast ratio between the characters and the background of the license plate. In this paper, edges of the vehicles are extracted by canny operator and topology structures of the characters are marked firstly, then candidate license plate areas are detected according to width height ratio and other characteristics after morphological operation. Finally, projection analysis is designed for accurate location of the license plate.